Machine learned coarse-grained protein force-fields: Are we there yet?

Area of Science:

• Computational chemistry
• Materials science
• Biophysics

Background:

• Machine learning (ML) is increasingly applied to scientific challenges, including developing accurate atomic-level force fields from quantum chemical data.
• ML-driven coarse-grained force fields are gaining importance for efficiently representing complex interactions and enabling simulations at larger scales.

Purpose of the Study:

• To review recent advancements in machine learning for coarse-grained force fields.
• To highlight ongoing efforts and challenges in developing transferable coarse-grained models.

Main Methods:

• Utilizing quantum chemical data to train ML models for atomic-level force fields.
• Developing ML approaches for coarse-grained force fields to capture higher-body interactions.
• Assessing the transferability of ML-derived coarse-grained models.

Main Results:

• Successful application of ML to create accurate atomic-level force fields.
• Growing relevance of ML for coarse-grained force fields to model omitted degrees of freedom.
• Significant challenges persist in achieving transferability for ML-based coarse-grained models.

Conclusions:

• ML shows great promise for both atomic and coarse-grained force fields.
• Further research is needed to overcome challenges in developing transferable ML coarse-grained models for large-scale simulations.